Date: Sat, 17 Jul 1999 17:18:40 -0700 (PDT) From: John Polstra <jdp@polstra.com> To: Doug Rabson <dfr@nlsystems.com> Cc: alpha@freebsd.org, Hidetoshi Shimokawa <simokawa@sat.t.u-tokyo.ac.jp> Subject: Re: alpha/12623: strtod(3) FPE on alphaev56 Message-ID: <XFMail.990717171840.jdp@polstra.com> In-Reply-To: <Pine.BSF.4.10.9907170925200.58023-100000@salmon.nlsystems.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Doug Rabson wrote:
> On Fri, 16 Jul 1999, Hidetoshi Shimokawa wrote:
>
>> A well-optimized program multiplying two 1000 * 1000 matrixes.
>>
>> % cc -mcpu=ev56 -O2 matrix2a.c
>> % time ./a.out
>> ./a.out 5.95s user 0.10s system 99% cpu 6.052 total
>> % cc -mieee -mcpu=ev56 -O2 matrix2a.c
>> % time ./a.out
>> ./a.out 21.86s user 0.08s system 99% cpu 21.976 total
>
> This is quite a severe penalty.
Yes, it is really bad!
> However, such code is hopefully fairly rare and the developer can
> still build the program using -mno-ieee for maximum performance so
> perhaps it isn't a problem?
It's hard to say. From what I've been told, fast FP is what many
people buy Alphas for. I'm starting to get cold feet about changing
the compiler default. Maybe sys.mk or bsd.{lib,prog}.mk would be
a better place to do it. Then at least it would be visible in the
"make" output.
> I really must build my 3D code on the alpha soon and see what kind of
> penalty I get for the maths in that.
Yes, it would be good to have another data point.
John
---
John Polstra jdp@polstra.com
John D. Polstra & Co., Inc. Seattle, Washington USA
"No matter how cynical I get, I just can't keep up." -- Nora Ephron
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-alpha" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?XFMail.990717171840.jdp>